gng module

class gng.GNG(iterations, learning_rate=0.1, age_threshold=100, node_creation_interval=100, new_node_error_discount=0.5, step_error_discount=0.1)

Bases: object

A class to apply the Growing-Neural-Gas algorithm over a given dataset, generating a Graph.

Parameters

  • iterations – The amount of iterations to run the GNG algorithm

  • learning_rate – Optional, default = 0.1 - The rate at which coordinates of nodes are updated towards the learned direction

  • age_threshold – Optional, default = 100 - The maximum age an edge can reach before being removed.

  • node_creation_interval – Optional, default = 100 - Defines at what amount of steps a new node should be added.

  • new_node_error_discount – Optional default = 0.5 - The discount value for the error when a new node is created.

  • step_error_discount – Optional default = 0.1 - The discount value for the error which is subtracted from the errors each step.

fit(data)

Fit the GNG class to the dataset.

Parameters

data – A list of coordinate tuples of data-points

Returns

This class instance.

transform(data)

Applies the self-organising-map algorithm over a given dataset and graph. Returns a graph after running the algorithm. NOTE: Running transform() multiple times will continue updating the same graph, to restart fresh create a new GNG instance.

Parameters

data – A list of coordinate tuples of data-points

Returns

The Graph of the GNG class after running the GNG algorithm.